Autonomous Vehicles
GM's Chief Product Officer Dismisses Humanoid Robots—and Explains Why the Automaker Bets on Software-First Development
Sterling Anderson argues that form-follows-function engineering, not humanoid hype, should drive factory automation and that personally owned autonomous vehicles offer a sounder path than robotaxis.
1/2
The Verge
General Motors' product chief Sterling Anderson sat down after the automaker's recent New York City showcase of its fresh technology roadmap—anchored on AI-driven design, advanced battery chemistry, integrated computing, and autonomous-driving systems—to discuss the strategic decisions steering the legacy Detroit manufacturer through an era of software-centric innovation and existential competitive threats.
The Core Bet: Smarter Vehicles, Faster Development#
At the heart of GM's overhaul lies a deceptively simple thesis: making the vehicle itself more intelligent. Anderson frames this shift not as a collection of isolated upgrades but as a unified "intelligence-defined vehicle" architecture that compresses typical development cycles—historically four to five years—down to roughly two years. The lever is virtualization. By running thousands of simulated design iterations rather than relying on a handful of physical prototypes sent to crash labs, engineering teams can validate safety and performance simultaneously, avoiding the false choice between speed and rigor.
The stakes are real. Chinese EV makers with razor-thin cost structures are flooding global markets. American EV demand has softened, forcing GM to reconcile massive battery-plant investments with ongoing software outlays. Betting the company's future on unproven AI systems and distributed architectures carries genuine risk of costly delays and missed transitions, especially against more nimble competitors like Tesla that were born digital.
Proprietary Data Plus Trusted Partnerships#
Anderson acknowledges that GM's scale—decades of design files, crash test data, and real-world vehicle telemetry—gives it a natural foundation for training foundation models. But transformer-based systems improve when co-trained with external datasets; the company uses third-party information across most AI implementations. The dividing line depends on the task: safety-critical functions (braking, steering, collision avoidance) run entirely onboard to eliminate latency and connectivity risk, while lower-stakes services like OnStar AI can tolerate cloud round-tripping.
Unifying Complexity, Preserving Brand Identity#
A century of multiband heritage—Chevrolet, GMC, Cadillac, Buick—historically fragmented engineering and slowed deployment. GM is imposing a common electrical and computing backbone across the entire portfolio, so software teams no longer juggle dozens of incompatible configurations. Yet aesthetic and experiential differentiation survives: the shares underlying powertrains, suspensions, and drivetrains across brands, while appearance, feature sets, and interface design remain brand-specific. Engineering becomes unified; brand identity remains skinned atop that shared substrate.

